Cleaning up systematic error in eye-tracking data by using required fixation locations

In the course of running an eye-tracking experiment, one computer system or subsystem typically presents the stimuli to the participant and records manual responses, and another collects the eye movement data, with little interaction between the two during the course of the experiment. This article demonstrates how the two systems can interact with each other to facilitate a richer set of experimental designs and applications and to produce more accurate eye tracking data. In an eye-tracking study, a participant is periodically instructed to look at specific screen locations, orexplicit required fixation locations (RFLs), in order to calibrate the eye tracker to the participant. The design of an experimental procedure will also often produce a number ofimplicit RFLs—screen locations that the participant must look at within a certain window of time or at a certain moment in order to successfully and correctly accomplish a task, but without explicit instructions to fixate those locations. In these windows of time or at these moments, the disparity between the fixations recorded by the eye tracker and the screen locations corresponding to implicit RFLs can be examined, and the results of the comparison can be used for a variety of purposes. This article shows how the disparity can be used to monitor the deterioration in the accuracy of the eye tracker calibration and to automatically invoke a re-calibration procedure when necessary. This article also demonstrates how the disparity will vary across screen regions and participants and how each participant’s uniqueerror signature can be used to reduce the systematic error in the eye movement data collected for that participant.     

Using machine learning to detect events in eye-tracking data

Event detection is a challenging stage in eye movement data analysis. A major drawback of current event detection methods is that parameters have to be adjusted based on eye movement data quality. Here we show that a fully automated classification of raw gaze samples as belonging to fixations, saccades, or other oculomotor events can be achieved using a machine-learning approach. Any already manually or algorithmically detected events can be used to train a classifier to produce similar classification of other data without the need for a user to set parameters. In this study, we explore the application of random forest machine-learning technique for the detection of fixations, saccades, and post-saccadic oscillations (PSOs). In an effort to show practical utility of the proposed method to the applications that employ eye movement classification algorithms, we provide an example where the method is employed in an eye movement-driven biometric application. We conclude that machine-learning techniques lead to superior detection compared to current state-of-the-art event detection algorithms and can reach the performance of manual coding.     

Face detection in complex visual displays: an eye-tracking study with 3- and 6-month-old infants and adults

The ability to detect and prefer a face when embedded in complex visual displays was investigated in 3- and 6-month-old infants, as well as in adults, through a modified version of the visual search paradigm and the recording of eye movements. Participants (N=43) were shown 32 visual displays that comprised a target face among 3 or 5 heterogeneous objects as distractors. Results demonstrated that faces captured and maintained adults' and 6-month-olds' attention, but not 3-month-olds' attention. Overall, the current study contributes to knowledge of the capacity of social stimuli to attract and maintain visual attention over other complex objects in young infants as well as in adults.     

Viewing distance and estimates of duration

28 students reproduced the durations of 12- and 18-sec. intervals, which were defined by a moving stimulus on a computer monitor. The monitor was viewed from a distance of either .4 or 2.5 m. Temporal reproductions were significantly greater for the far than for the near distance conditions. Results are considered from the perspective of the change models of timing.     

The control of fixation duration in visual search

A study is reported how fixation durations are affected as observers voluntarily use information in the visual environment to direct saccades in search of a target. In experiments with direction-coded displays, search fixation durations for correct and incorrectly directed saccades were comparable during image-driven/bottom-up search. In contrast, for knowledge-driven search, the relationship depended on peripheral preview. With ample access to peripheral preview, search fixation durations preceding correct saccades were shorter than those preceding incorrectly directed saccades. However, they were longer when preview was constrained by a gaze-contingent moving window. Fixation control was modeled as sigmoid functions.     

A hidden markov model-based analysis framework using eye-tracking data to characterise re-orientation strategies in minimally invasive surgery

Natural Orifice Translumenal Endoscopic Surgery (NOTES) represents a new paradigm in minimally invasive surgery whereby an operator uses a flexible instrument to navigate within an anatomical space. The flexible instrument is introduced via a natural body orifice such as the mouth to perform an operative intervention controlled by the surgeon without making incisions in the anterior abdominal wall. It differs from laparoscopic (traditional keyhole) surgery in two main ways: (1) the instruments used in NOTES are not rigid as in laparoscopic surgery and not specifically designed for this application; and (2) there are no incisions in the anterior abdominal wall. Disorientation during NOTES is a significant problem because as in laparoscopic surgery the operator typically uses a 2D camera-monitor interface for visualisation by the camera at the site of the operation. Furthermore, the positional cue offered by the external component (outside the body) of the camera is absent, and camera navigation is more cumbersome. When the operator is disorientated, it is important to be able to re-orientate quickly to minimise potential surgical errors. It is hypothesised that when surgeons become disorientated, there exist discrete patterns in psychophysical behaviour which are associated with effective re-orientation, and that these patterns are recognisable. In this study, we examine visual re-orientation behaviour in 18 subjects using eye-tracker data in a model comprised of selective image manipulation of everyday objects in a box trainer. We characterise effective behaviour using a fixation sequence similarity-based hidden Markov model. We show that the output of this algorithm is reliable in differentiating visual behavioural sequences, and that there are specific behavioural patterns and strategies associated with successful re-orientation in this model. Good re-orientation strategy appears to rely on identification and focus on a central object within the scene and judging position of its surrounding peripheral objects, suggesting integration of both geometric and feature information in a systematic way. Using selective, inconsistent feature cues for re-orientation were associated with less effective performance.     

Development of binocular fixation in human infants

Two experiments measured changes in binocular eye alignment from 1- to 6-month-old human infants. In Experiment 1 changes in binocular eye alignment were recorded from 1, 2, and 3 month olds using corneal photography. A luminous target was moved along the infant's midline at one of two constant speeds. Infants at all three ages showed some evidence of appropriate changes in binocular eye alignment (convergence as the target approached and divergence as the target receded). The likelihood of appropriate convergence and divergence increased with age, as did the ability to respond appropriately to the faster target motion. A measure of convergence lag (further decrease in interpupillary distance after target motion had ended) was obtained for trials on which the target approached the infant. Convergence lag decreased with age and was greater on trials at the faster target speed until 3 months of age. Experiment 2 measured the responses of 3, $\text{4}\text{1}\text{2}$, and 6 month olds as a wedge prism was placed alternately in front of each eye. The prism displaced the image nasally (either 2.5° or 5°), creating inappropriate binocular eye alignment. A refixation eye movement in response to the prism was not present consistently until 6 months of age. Hypotheses as to the mechanism controlling infant binocular eye alignment are discussed with respect both to the present findings and to findings from human adults and animals.     

Control of fixation duration in a simple search task

To obtain insight into the control of fixation duration during visual search, we had 4 subjects perform simple search tasks in which we systematically varied the discriminability of the target. The experiment was carried out under two conditions. Under the first condition (blocked), the discriminability of the target was kept constant during a session. Under the second condition (mixed), the discriminability of the target varied per trial. Under the blocked condition, fixation duration increased with decreasing discriminability. For 2 subjects, we found much shorter fixation durations in difficult trials with the mixed condition than in difficult trials with the blocked condition. Overall, the subjects fixated the target, continued to search, and then went back to the target in 5%–55% of the correct trials. In these trials, the result of the analysis of the foveal target was not used for preparing the next saccade. The results support a preprogramming model of the control of fixation duration. In a simple search task, control of fixation duration appears to be indirect.     

The relationship between estimates of foreperiod duration and simple time reaction

The relationship between estimates of foreperiod length and simple reaction time (RT) was examined using constant foreperiods. It was learned that neither the average accuracy nor the variability of the foreperiod estimates necessarily manifested themselves in appropriate simple RT change. Of secondary interest was the discovery that the average accuracy of foreperiod duration estimations differed greatly from that obtained for similar time lengths in a time-estimation task. Finally, an attempt was made to determine whether subjects actually utilized the directional knowledge of results inherent in the imperative stimulus to influence their foreperiod duration judgments. Unfortunately, the evidence obtained did not permit a definite answer regarding this possibility.     

Effect of learning on production of duration in variable motor conditions

It has been suggested that increasing proprioceptive feedback and ensuring its consistency from trial to trial favours timing accuracy. The progressive acquisition of a timing performance with isometric and anisometric responses was investigated in conditions of 'inconsistent feedback', with two different methods: counting seconds or not. Subjects gripped the handle of a dynamometer during precisely 5 seconds in 4 randomly distributed conditions: 'Weak' or 'Strong' constant force, 'Slow' or 'Rapid' variable force. A first group produced the durations without counting and a second group counted seconds either mentally or aloud. Learning was examined in 4 successive blocks of trials. Average produced durations did not differ as a function of group, but the variability was reduced when subjects counted seconds. In both groups, the constant force conditions induced more accurate responses than the variable force conditions in the first block of trials. 'Slow' and 'Rapid' conditions were respectively associated with overestimation and under-estimation of response duration. These trends diminished progressively with learning. Both groups yielded sequential effects linked to duration and independent of condition. These data suggest that, whether subjects counted or not, learning was not based on condition-specific proprioceptive feedback. It may involve the elaboration of an internal temporal reference common to all conditions.     

Color preferences in infants and adults are different

Adults commonly prefer blues most and greenish yellows least, but these hue preferences interact with lightness and saturation (e.g., dark yellow is particularly disliked: Palmer & Schloss (Proceedings of the National Academy of Sciences 107:8877–8882, 2010)). Here, we tested for a similar hue-by-lightness interaction in infant looking preferences, to determine whether adult preferences are evident early in life. We measured looking times for both infants and adults in the same paired-comparison task using all possible pairs of eight colors: four hues (red/yellow/green/blue) at two lightness levels (dark/light). The adult looking data were strikingly similar to other adults’ explicit preference responses, indicating for the first time that adults look longer at colors that they like. Infants showed a significant hue-by-lightness interaction, but it was quite different from the adult pattern. In particular, infants had a stronger looking preference for dark yellow and a weaker preference for light blue than did adults. The findings are discussed in relation to theories on the origins of color preference.     

Solving the chicken-and-egg problem of letter detection and fixation duration in reading

When asked to detect target letters while reading a text, participants miss more letters in frequent function words than in less frequent content words. According to the truncation assumption that characterizes most models of this effect, misses occur when word-processing time is shorter than letter-processing time. Fixation durations for detections and omissions were compared with fixation durations from a baseline condition when participants were searching for a target letter embedded in different words. Although, as predicted by truncation, fixation durations were longer for detections than for omissions, fixation durations for detections were also longer than those for the same words in the baseline condition, demonstrating that longer fixation durations when targets are detected are more likely to be due to demands associated with producing a detection response than to truncation. Also, contrary to predictions from the truncation assumption, the standard deviation of fixation durations for detections was larger than that from the baseline condition.     

Search asymmetry and eye movements in infants and adults

Search asymmetry is characterized by the detection of a feature-present target amidst feature-absent distractors being efficient and unaffected by the number of distractors, whereas detection of a feature-absent target amidst feature-present distractors is typically inefficient and affected by the number of distractors. Although studies have attempted to investigate this phenomenon with infants (e.g., Adler, Inslicht, Rovee-Collier, & Gerhardstein in Infant Behavioral Development, 21, 253–272, 1998; Colombo, Mitchell, Coldren, & Atwater in Journal of Experimental Psychology: Learning, Memory and Cognition, 19, 98–109, 1990), due to methodological limitations, their findings have been unable to definitively establish the development of visual search mechanisms in infants. The present study assessed eye movements as a means to examine an asymmetry in responding to feature-present versus feature-absent targets in 3-month-olds, relative to adults. Saccade latencies to localize a target (or a distractor, as in the homogeneous conditions) were measured as infants and adults randomly viewed feature-present (R among Ps), feature-absent (P among Rs), and homogeneous (either all Rs or all Ps) arrays at set sizes of 1, 3, 5, and 8. Results indicated that neither infants’ nor adults’ saccade latencies to localize the target in the feature-present arrays were affected by increasing set sizes, suggesting that localization of the target was efficient. In contrast, saccade latencies to localize the target in the feature-absent arrays increased with increasing set sizes for both infants and adults, suggesting an inefficient localization. These findings indicate that infants exhibit an asymmetry consistent with that found with adults, providing support for functional bottom-up selective attention mechanisms in early infancy.     

Control of fixation duration in visual search and memory search: another look

In visual search a variable delay (up to 150 msec) between the beginning of each fixation and the onset of a search stimulus reduces the time (oculomotor latency) between stimulus onset and the subject's next saccadic eye movement. Two hypotheses for this effect of stimulus onset delay (SOD) were compared: first, process monitoring, that SOD simply serves as a warning interval to facilitate saccadic responses; and second, preprogramming, that saccades are preprogrammed at short SODs. In the first experiment SOD produced a decline in oculomotor latency in search similar to that seen in previous studies. In the second and third experiments, the size of the memory set in a Sternberg memory search paradigm was varied, or a mask flanking some of the search stimuli was used, to vary the processing time of each stimulus. Partial preprogramming of saccades at short delays would predict that increasing the processing time of individual stimuli would increase oculomotor latency at only short SODs. However, oculomotor latency increased equally at all SODs. In this search task, then, the SODs appeared to facilitate saccade initiation.     

No evidence for accurate visuomotor memory: systematic and variable error in memory-guided reaching

The authors explored whether the motor system has access to highly accurate information about the aiming environment after visual occlusion. Participants (N = 14) reached to 1 of 3 midsagittal targets in 4 visual conditions (open-loop, brief-delay, 500-ms delay, and 2,000-ms delay). In all conditions, the aiming environment was first viewed for 2,000 ms. Movements were cued immediately after the initial viewing period in the open-loop and brief-delay conditions. Vision was not occluded until movement onset in the open-loop condition, whereas vision was occluded coincidentally with the movement cue in the brief-delay condition. In the 2 longer delay conditions, the movement was cued following a 500- or a 2,000-ms no-vision delay period. Participants overshot the target in the open-loop condition, but that tendency was significantly reduced in the 3 delay conditions. Moreover, endpoint variability was greater in the 3 delay conditions than in the open-loop condition. A speed-accuracy tradeoff account could not explain the differences between open-loop and delayed reaching. Those findings suggest that the motor system does not have access to highly accurate information about the aiming environment for any appreciable period of time following visual occlusion, consistent with the view that the visuomotor system operates in real time.     

Perception of the duration of auditory and visual stimuli in children and adults

This experiment investigated the effect of modality on temporal discrimination in children aged 5 and 8 years and adults using a bisection task with visual and auditory stimuli ranging from 200 to 800 ms. In the first session, participants were required to compare stimulus durations with standard durations presented in the same modality (within-modality session), and in the second session in different modalities (cross-modal session). Psychophysical functions were orderly in all age groups, with the proportion of long responses (judgement that a duration was more similar to the long than to the short standard) increasing with the stimulus duration, although functions were flatter in the 5-year-olds than in the 8-year-olds and adults. Auditory stimuli were judged to be longer than visual stimuli in all age groups. The statistical results and a theoretical model suggested that this modality effect was due to differences in the pacemaker speed of the internal clock. The 5-year-olds also judged visual stimuli as more variable than auditory ones, indicating that their temporal sensitivity was lower in the visual than in the auditory modality.     

Perceived tachistoscopic duration and early visual processing in preschool children and adults

Two experiments further explored the Avant et al. (1975) finding that stimulus familiarity influences prerecognition processing to generate differences in the apparent duration of tachistoscopic flashes. The first experiment tested for developmental differences in the effects of upright versus 90°, 180°, and 270° rotations of a single letter or number upon the apparent duration of pre- and postmasked 30- and 50-msec flashes with adults and 4- and 5-year-old children. All age groups judged upright presentations to be of briefer duration. These differences in apparent duration were interpreted to index the automaticity of contacts between stimulus inputs and their memory representations. Failure of the children to recognize the letter and number in any orientation indicates that contact between stimulus inputs and memory representations precedes allocation of attention to the presented stimulus. The second experiment explored the influence of spatial structures which are not coded verbally by testing effects of good and poor dot pattern Gestalts on the apparent duration of tachistoscopic flashes. Adults discriminated between apparent durations of good and poor Gestalts but 4- and 5-year-olds did not. Apparent duration differences in the two experiments showed that spatial pattern structure and familiarity with verbal stimuli influence early visual processing in different ways.     

A comparison of classical eyelid conditioning in adults and infants

In a classical conditioning procedure an eyeblink-eliciting tap to the glabella (the flat region of skin between the eyebrows) was presented 500 ms after the onset of a mild l-kHz tone. As tone-tap presentations proceeded, the probability of an eyeblink during the latter part of the tone increased in both infants (median age 8 months) and adults, but the infants were slower to condition than the adults and were more variable. Overall, the latency of the conditioned response to tone was significantly longer for infants than for adults, but the latency of the unconditioned response to tap was significantly shorter for infants than for adults.